Monday, August 9, 2010

What is Statistical Validation?

I used to teach SPC (Statistical Process Controls). I don’t think I ever did it very well, and one of the troubles was that it seemed that most of the students locked up and their eyes glassed over at the first mention of the “S word.” Try as we may, our group just could never get the detailed math to be successfully absorbed by the hourly and salaried employees of our manufacturing organization. Transferring the knowledge of the math of standard deviations, chi squared distributions, and curve skewness was a lost cause to the overwhelming majority of our 8th grade education level hourly employees, and not a lot better with our college level professionals.


That led us to develop and train a different kind of SPC, Simple Process Controls. Our tools became non mathematical constructs like Pareto Charts, Process Maps, Cause and Effect Diagrams and Kaizen teams. In other words a simple, practical, non mathematical approach that helped our teams to solve well over 90% of the day-to-day problems in an industrial organization. This approach was incredibly successful world wide. We weren’t worried about statistical validation of the data base; rather it was a focus on permanently fixing what was in front of our eyes.


Yet there were certain applications that did require statistical rigor. And what does that mean without using the math language of statisticians? It means that there is a level of certainty that what is done will lead to a desired result. In the statistics world they call this a correlation factor. If you follow a certain detailed process you will likely get a high probability of achieving the desired result.


One of my favorite examples comes from the late safety culture giant Dr. Dan Petersen. I can still hear him say something like “Why do you pay attention to setting goals based on injury rates, something you do not want to have happen? The numbers are statistically invalid and it is just plain illogical and ridiculous to count things you don’t want to occur!” Statistically invalid? Yes, we are dealing at the supervisory level where there aren’t enough hours in a crew to do other than cause a supervisor to fail if they have even one injury in a multiple year time frame. Their total Case Rate is based on 100 effort years and one injury is impossible to overcome. It’s not much better at the facility level.


Is there such a thing as a statistically valid measure of safety performance? Dr. Dan and I agreed, count the things you want to occur, set goals for them and reward excellence in performance. This is not based on injury rates. The measure is the accomplishment of well thought out safety accountability actions that must be repeated correctly innumerable times. If something goes wrong, an incident or a near miss, form a team, fix the root causes and improve the safety accountabilities that will forever eliminate the possibility of this error recurring.


Stop counting what you don’t want to occur and reward what you do want. When your safety culture has this kind of focus the statistics will take care of themselves.


The Doc

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.

Connect With Us

Bookmark and Share
/////////////Google analytics tracking script//////////////// /////////////END -- Google analytics tracking script////////////////